HELP! - My MERGE Statement Has More Than One Data Set With Repeats of BY Values!

Transcription

1 HELP! - My MERGE Statement Has More Than One Data Set With Repeats of BY Values! Andrew T. Kuligowski, The Nielsen Company ABSTRACT Most users of the SAS system have encountered the following message: NOTE: MERGE statement has more than one data set with repeats of BY values. There are many papers in the Proceedings from past SAS User Group conferences that describe how to programatically force this "many-to-many" merge to occur. These presentations are built upon the assumption that the user wants to merge datasets that have repeats of BY values. However, it is also possible that the coder did not expect this condition and wishes to avoid and eliminate it. This presentation will demonstrate a technique using a series of ad hoc programs that will isolate the observations causing the many-to-many merge, so that we can modify our underlying data assumptions, and hopefully eliminate this condition. MERGE Statement: Repeats of BY Values As previously mentioned, most users of the SAS system have encountered the following message: NOTE: MERGE statement has more than one data set with repeats of BY values. This implies an error in your SAS routine, caused by a misunderstanding of the input data. We want to isolate these cases and modify our assumptions, so that we can correct our MERGE and eliminate this condition. The following will illustrate an example of "repeats of BY values". We are going to merge a dataset containing a list of dog breeds [see Table "1-A"] against a dataset containing the dogs owned by sample households [see Table "1-B"] using the variable BREED, keeping only those records from the Breed file that have a corresponding record in the Household file. As one would expect (knowing, of course, that this example was created to illustrate the problem at hand), the merge results in "more than one data set with repeats of BY values" [illustrated in Table "1-C"]. The problem is to determine which BY values had repeats, which records (on which files) are affected, and what additional information needs to be included to make the "BY values", also referred to as "merge variables", unique on at least one of the files. BREED VARIETY OTHER INFO Bulldog Dalmatian Dachshund Mini Dachshund MiniLonghair Dachshund MiniWirehair Dachshund Std BREED VARIETY OTHER INFO Dachshund Std Longhair Dachshund Std Wirehair Ger Sheprd GoldRetrvr Husky,Sib Lab Retrvr Table "1-A" : Breed File 1

2 HHLD ID BREED VARIETY BIRTHDAY GOAWAYDY Dalmatian 07/31/ Dalmatian 12/23/ Bulldog English 05/19/91 02/20/ Dachshund Std Longhair 09/17/ Dachshund Std Longhair 10/29/ Ger Sheprd 11/24/89 12/07/ Husky,Sib 05/26/ Lab Retrvr 02/28/94 05/06/ GoldRetrvr 03/06/95 Table "1-B" : Dogs per Household File 177 DATA DOGDTL; 178 MERGE DOGOWNED (IN=IN_OWNED) 179 DOGBREED (IN=IN_BREED); 180 BY BREED ; 181 IF IN_OWNED ; 182 RUN; NOTE: MERGE statement has more than one data set with repeats of BY values. NOTE: The data set WORK.DOGDTL has 13 observations and 6 variables. Table "1-C" : SASLOG - MERGE with "repeats of BY values" We can use basic elements of the SAS System to do most of the analysis for us; our most powerful tool will be the MEANS procedure. PROC MEANS is traditionally used to compute descriptive statistics for numeric variables in a SAS Dataset. In this situation, we simply need a list of the unique values of our merge variable (or, in a more complex case, the unique combinations of values for our merge variables), along with a count of the number of occurrences of each in both of our datasets. The NWAY option on the PROC MEANS statement will limit the output dataset to only those observations with the "highest interaction among CLASS variables" -- that is, only those records containing the unique values or combinations of values for the BY variables will be written to the output dataset. NOPRINT is optional; it can be used or excluded based on personal preference. The variable used in the VAR statement can be any numeric variable in the dataset; the only condition is that it must be a numeric variable. Date variables and ID values should be considered, since many / most datasets contain them and they are normally stored as numeric values. (In the rare event your dataset contains exclusively character variables, you will need to add a numeric variable to either the original dataset or a copy of it prior to issuing PROC MEANS.) The OUTPUT statement must specify an OUT= dataset. It must also include the statistics keyword N=, so that a record count -- and only a record count -- for each unique combination of values for the BY variables will be written to each observation of the output dataset. [Table "1-D" illustrates this use of PROC.] The next step is to take the outputs of our PROC MEANS for each affected dataset and merge them together. We cannot encounter the "... repeats of BY values" note, since we now only have one observation per unique combination of BY values! Each observation does have a count of the number of observations that contain the combination of BY values in their original dataset, stored as _FREQ_ by default. The RENAME= option the datasets in the MERGE statement can be used to give the _FREQ_ variable in each dataset a unique name in our output dataset. (Alternatively, a clearner approach would be to change the N= option on the OUTPUT statement in PROC MEANS to N=varname, avoiding the need for the subsequent RENAME.) We will use a subsetting IF, so that we only keep those observations that have multiple occurrences in each input dataset. The output of this step will contain the unique combinations of values that are causing the "...repeats of BY values" note. [The routine is contained in Table "1-E", and the output depicted in Table "1-F".] 2

3 208 PROC MEANS DATA=DOGOWNED NOPRINT NWAY; 209 CLASS BREED ; 210 VAR HHLD_ID ; 211 OUTPUT OUT=SUMOWNED N=CNTOWNED; 212 RUN; NOTE: The data set WORK.SUMOWNED has 7 observations and 4 variables. NOTE: The PROCEDURE MEANS used 0.05 seconds. 213 PROC MEANS DATA=DOGBREED NOPRINT NWAY; 214 CLASS BREED ; 215 VAR OTHRINFO ; 216 OUTPUT OUT=SUMBREED N=; 217 RUN; NOTE: The data set WORK.SUMBREED has 7 observations and 4 variables. NOTE: The PROCEDURE MEANS used 0.05 seconds. Table "1-D" : SASLOG - PROC MEANS example 585 DATA SUMMERGE (KEEP=BREED CNTBREED CNTOWNED); 586 MERGE SUMBREED (RENAME=(_FREQ_=CNTBREED)) 587 SUMOWNED ; 588 BY BREED ; 589 IF CNTBREED > 1 AND CNTOWNED > 1; 590 RUN ; NOTE: The data set WORK.SUMMERGE has 1 observations and 3 variables. Table "1-E" : SASLOG - Merging the PROC MEANS outputs SAS Dataset WORK.SUMMERGE OBS BREED CNTBREED CNTOWNED 1 Dachshund 6 2 Table "1-F" : Results of Merging the PROC MEANS outputs Up to this point, we have not discussed one important factor in this analysis - the human element. The process described in this section is meant to be used as a tool to guide the analyst through the unknown elements in their data - once these areas become known - there is no need to continue this analysis. The listing of unique combinations of values that occur multiple times in each dataset will often be that stopping point for an analyst. The information obtained will allow them to make modifications to their assumptions and corresponding changes to their routines. However, in the event that the oddities are still not clear, we can employ one or more additional MERGE steps, taking the merged outputs from the PROC MEANS, and merging those dataset against each of the original datasets. [Table "1-G" shows how this is done, while Table 1-H displays the results of this analysis.] This final step should provide sufficient clarification for the analyst to determine which factor(s) are missing in their assumptions and adjust their routines accordingly. 3

4 599 DATA CHKOWNED ; 600 MERGE DOGOWNED (IN=IN_BREED ) 601 SUMMERGE (IN=IN_MERGE ); 602 BY BREED ; 603 IF IN_MERGE ; 604 RUN ; NOTE: The data set WORK.CHKOWNED has 2 observations and 7 variables. 605 DATA CHKBREED ; 606 MERGE DOGBREED (IN=IN_BREED ) 607 SUMMERGE (IN=IN_MERGE ); 608 BY BREED ; 609 IF IN_MERGE ; 610 RUN ; NOTE: The data set WORK.CHKBREED has 6 observations and 5 variables. Table "1-G" : SASLOG - Merging the analysis back to the original input SAS Dataset WORK.CHKOWNED OBS HHLD_ID BREED VARIETY BIRTHDAY GOTCHADY CNTBREED CNTOWNED Dachshund Std Longhair Dachshund Std Longhair SAS Dataset WORK.CHKBREED OBS BREED VARIETY OTHRINFO CNTBREED CNTOWNED 1 Dachshund Mini Dachshund MiniLonghair Dachshund MiniWirehair Dachshund Std Dachshund Std Longhair Dachshund Std Wirehair Table "1-H" : Results of merging the analysis back to the original input The end result of this analysis is the discovery that the BY statement in the routine does not contain the proper variables to uniquely identify each record. By adding the extra variable or variables to the original BY statement, the routine works without error and without the offending NOTE. [Table "1-I" shows the corrected MERGE statement.] 682 DATA DOGDTAIL; 683 MERGE DOGOWNED (IN=IN_OWNED) 684 DOGBREED (IN=IN_BREED); 685 BY BREED VARIETY; 686 IF IN_OWNED ; 687 RUN; NOTE: The data set WORK.DOGDTAIL has 9 observations and 6 variables. Table "1-I" : SASLOG - MERGE without "repeats of BY values" 4

5 CONCLUSION This paper addressed the use of ad hoc routines to explore WHY a common SASLOG message occurred, and covered how to correct a routine to prevent the recurrence of that message. It is hoped that the mechanisms discussed in this paper might be used by the readers in their daily jobs. However, this paper is a failure -- at least in part -- if the process stops there. It is hoped, even more strongly, that the concepts of developing and using ad hoc routines to fully understand ones data are the true lessons that the reader retains from this paper. REFERENCES / FOR FURTHER INFORMATION Burlew, Michele M. (2001). Debugging SAS Programs A Handbook of Tools and Techniques. Cary, NC: SAS Institute, Inc. Cody, Ron (2008). Cody s Data Cleaning Techniques Using SAS, Second Edition. Cary, NC: SAS Institute, Inc. Kuligowski, Andrew T. (2003), "The BEST. Message in the SASLOG". Proceedings of the Twenty- Eighth Annual SAS Users Group International Conference. Cary, NC: SAS Institute, Inc. SAS Institute, Inc. (2008), SAS 9.2 Online Documentation. Cary, NC: SAS Institute, Inc. SAS Institute, Inc. (1990), SAS Language: Reference, Version 6, First Edition. Cary, NC: SAS Institute, Inc. SAS Institute, Inc. (2007), SAS OnlineDoc, Cary, NC: SAS Institute, Inc. SAS and all other SAS Institute Inc. product or service names are registered trademarks or trademarks of SAS Institute Inc. in the USA and other countries. indicates USA registration. The author can be contacted via at: A_Kuligowski@msn.com (preferred) or Andy.Kuligowski@Nielsen.com ACKNOWLEDGMENTS The author wishes to acknowledge those individuals, whose identities have been lost to the ages, whose data oddities across the years provided the true inspiration behind this paper. 5